import gradio as gr
from transformers import MistralForCausalLM, AutoTokenizer
import torch

model_name = "sand-ai/MAGI-1"

tokenizer = AutoTokenizer.from_pretrained(model_name)
model = MistralForCausalLM.from_pretrained(
    model_name,
    torch_dtype=torch.float32,
    device_map="cpu"
)

def chat_fn(message, history):
    inputs = tokenizer(message, return_tensors="pt")
    outputs = model.generate(**inputs, max_new_tokens=200)
    reply = tokenizer.decode(outputs[0], skip_special_tokens=True)
    return reply

demo = gr.ChatInterface(chat_fn)

demo.launch()